README.md

---
language: sl
language_name: Slovenian
language_family: slavic_south
tags:
  - wikilangs
  - nlp
  - tokenizer
  - embeddings
  - n-gram
  - markov
  - wikipedia
  - feature-extraction
  - sentence-similarity
  - tokenization
  - n-grams
  - markov-chain
  - text-mining
  - fasttext
  - babelvec
  - vocabulous
  - vocabulary
  - monolingual
  - family-slavic_south
license: mit
library_name: wikilangs
pipeline_tag: text-generation
datasets:
  - omarkamali/wikipedia-monthly
dataset_info:
  name: wikipedia-monthly
  description: Monthly snapshots of Wikipedia articles across 300+ languages
metrics:
  - name: best_compression_ratio
    type: compression
    value: 4.308
  - name: best_isotropy
    type: isotropy
    value: 0.7907
  - name: vocabulary_size
    type: vocab
    value: 0
generated: 2026-01-17
---

# Slovenian - Wikilangs Models
## Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on **Slovenian** Wikipedia data.
We analyze tokenizers, n-gram models, Markov chains, vocabulary statistics, and word embeddings.

## 📋 Repository Contents

### Models & Assets

- Tokenizers (8k, 16k, 32k, 64k)
- N-gram models (2, 3, 4, 5-gram)
- Markov chains (context of 1, 2, 3, 4 and 5)
- Subword N-gram and Markov chains
- Embeddings in various sizes and dimensions (aligned and unaligned)
- Language Vocabulary
- Language Statistics

![Performance Dashboard](visualizations/performance_dashboard.png)

### Analysis and Evaluation

- [1. Tokenizer Evaluation](#1-tokenizer-evaluation)
- [2. N-gram Model Evaluation](#2-n-gram-model-evaluation)
- [3. Markov Chain Evaluation](#3-markov-chain-evaluation)
- [4. Vocabulary Analysis](#4-vocabulary-analysis)
- [5. Word Embeddings Evaluation](#5-word-embeddings-evaluation)
- [6. Morphological Analysis (Experimental)](#6--morphological-analysis-experimental)
- [7. Summary & Recommendations](#7-summary--recommendations)
- [Metrics Glossary](#appendix-metrics-glossary--interpretation-guide)
- [Visualizations Index](#visualizations-index)

---
## 1. Tokenizer Evaluation

![Tokenizer Compression](visualizations/tokenizer_compression.png)

![Tokenizer Fertility](visualizations/tokenizer_fertility.png)

![Tokenizer OOV](visualizations/tokenizer_oov.png)

![Total Tokens](visualizations/tokenizer_total_tokens.png)

### Results

| Vocab Size | Compression | Avg Token Len | UNK Rate | Total Tokens |
|------------|-------------|---------------|----------|--------------|
| **8k** | 3.326x | 3.33 | 0.6705% | 1,060,027 |
| **16k** | 3.677x | 3.68 | 0.7414% | 958,786 |
| **32k** | 4.017x | 4.02 | 0.8100% | 877,496 |
| **64k** | 4.308x 🏆 | 4.31 | 0.8686% | 818,369 |

### Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

**Sample 1:** `Tinja je lahko: Tinja Donja (Bosna in Hercegovina) Tinja Gornja (Bosna in Herceg...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁tin ja ▁je ▁lahko : ▁tin ja ▁do nja ▁( ... (+27 more)` | 37 |
| 16k | `▁tin ja ▁je ▁lahko : ▁tin ja ▁donja ▁( bosna ... (+22 more)` | 32 |
| 32k | `▁tin ja ▁je ▁lahko : ▁tin ja ▁donja ▁( bosna ... (+21 more)` | 31 |
| 64k | `▁tin ja ▁je ▁lahko : ▁tin ja ▁donja ▁( bosna ... (+21 more)` | 31 |

**Sample 2:** `Škrabčeva ulica je lahko naziv več ulic: Škrabčeva ulica, Ljubljana Škrabčeva ul...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁š kra b čeva ▁ulica ▁je ▁lahko ▁naziv ▁več ▁u ... (+17 more)` | 27 |
| 16k | `▁š kra b čeva ▁ulica ▁je ▁lahko ▁naziv ▁več ▁ulic ... (+16 more)` | 26 |
| 32k | `▁škra b čeva ▁ulica ▁je ▁lahko ▁naziv ▁več ▁ulic : ... (+13 more)` | 23 |
| 64k | `▁škrab čeva ▁ulica ▁je ▁lahko ▁naziv ▁več ▁ulic : ▁škrab ... (+10 more)` | 20 |

**Sample 3:** `Kreševo je lahko: Kreševo, Bosna in Hercegovina Kreševo, Hrvaška glej tudi Kruše...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁kre še vo ▁je ▁lahko : ▁kre še vo , ... (+13 more)` | 23 |
| 16k | `▁kre še vo ▁je ▁lahko : ▁kre še vo , ... (+13 more)` | 23 |
| 32k | `▁kre ševo ▁je ▁lahko : ▁kre ševo , ▁bosna ▁in ... (+9 more)` | 19 |
| 64k | `▁kre ševo ▁je ▁lahko : ▁kre ševo , ▁bosna ▁in ... (+9 more)` | 19 |


### Key Findings

- **Best Compression:** 64k achieves 4.308x compression
- **Lowest UNK Rate:** 8k with 0.6705% unknown tokens
- **Trade-off:** Larger vocabularies improve compression but increase model size
- **Recommendation:** 32k vocabulary provides optimal balance for production use

---
## 2. N-gram Model Evaluation

![N-gram Perplexity](visualizations/ngram_perplexity.png)

![N-gram Unique](visualizations/ngram_unique.png)

![N-gram Coverage](visualizations/ngram_coverage.png)

### Results

| N-gram | Variant | Perplexity | Entropy | Unique N-grams | Top-100 Coverage | Top-1000 Coverage |
|--------|---------|------------|---------|----------------|------------------|-------------------|
| **2-gram** | Word | 230,065 | 17.81 | 1,580,681 | 8.6% | 18.2% |
| **2-gram** | Subword | 307 🏆 | 8.26 | 19,173 | 64.1% | 99.0% |
| **3-gram** | Word | 628,655 | 19.26 | 2,417,119 | 4.1% | 11.2% |
| **3-gram** | Subword | 3,012 | 11.56 | 154,341 | 21.4% | 65.5% |
| **4-gram** | Word | 1,262,619 | 20.27 | 3,613,803 | 3.4% | 8.8% |
| **4-gram** | Subword | 20,514 | 14.32 | 898,859 | 9.3% | 30.9% |
| **5-gram** | Word | 764,287 | 19.54 | 2,274,837 | 4.9% | 11.6% |
| **5-gram** | Subword | 99,057 | 16.60 | 3,187,233 | 4.7% | 16.9% |

### Top 5 N-grams by Size

**2-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `je bil` | 242,649 |
| 2 | `se je` | 241,005 |
| 3 | `ki je` | 174,450 |
| 4 | `je bila` | 164,565 |
| 5 | `ki so` | 110,725 |

**3-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `glej tudi seznam` | 56,792 |
| 2 | `pr n št` | 34,003 |
| 3 | `ki ga je` | 25,570 |
| 4 | `sklici zunanje povezave` | 21,261 |
| 5 | `opombe glej tudi` | 21,007 |

**4-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `glej tudi seznam naselij` | 20,688 |
| 2 | `in opombe glej tudi` | 20,455 |
| 3 | `opombe glej tudi seznam` | 19,924 |
| 4 | `viri in opombe glej` | 16,449 |
| 5 | `ki upravno spada pod` | 15,090 |

**5-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `in opombe glej tudi seznam` | 19,642 |
| 2 | `opombe glej tudi seznam naselij` | 16,702 |
| 3 | `viri in opombe glej tudi` | 16,449 |
| 4 | `glej tudi seznam naselij v` | 10,861 |
| 5 | `glej tudi seznam naselij na` | 9,825 |

**2-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `a _` | 11,379,031 |
| 2 | `e _` | 9,711,848 |
| 3 | `i _` | 7,963,628 |
| 4 | `_ p` | 7,248,467 |
| 5 | `_ s` | 7,019,681 |

**3-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `j e _` | 4,141,312 |
| 2 | `_ j e` | 2,970,998 |
| 3 | `_ p o` | 2,883,740 |
| 4 | `_ p r` | 2,720,662 |
| 5 | `_ n a` | 2,547,716 |

**4-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ j e _` | 2,761,955 |
| 2 | `_ i n _` | 2,026,152 |
| 3 | `_ n a _` | 1,018,082 |
| 4 | `_ p r e` | 991,101 |
| 5 | `e g a _` | 882,400 |

**5-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `a _ j e _` | 682,064 |
| 2 | `, _ k i _` | 656,040 |
| 3 | `_ l e t a` | 563,205 |
| 4 | `_ j e _ b` | 529,211 |
| 5 | `j e _ b i` | 504,001 |


### Key Findings

- **Best Perplexity:** 2-gram (subword) with 307
- **Entropy Trend:** Decreases with larger n-grams (more predictable)
- **Coverage:** Top-1000 patterns cover ~17% of corpus
- **Recommendation:** 4-gram or 5-gram for best predictive performance

---
## 3. Markov Chain Evaluation

![Markov Entropy](visualizations/markov_entropy.png)

![Markov Contexts](visualizations/markov_contexts.png)

![Markov Branching](visualizations/markov_branching.png)

### Results

| Context | Variant | Avg Entropy | Perplexity | Branching Factor | Unique Contexts | Predictability |
|---------|---------|-------------|------------|------------------|-----------------|----------------|
| **1** | Word | 1.0363 | 2.051 | 12.35 | 1,898,880 | 0.0% |
| **1** | Subword | 0.7016 | 1.626 | 5.31 | 18,900 | 29.8% |
| **2** | Word | 0.3404 | 1.266 | 2.08 | 23,427,544 | 66.0% |
| **2** | Subword | 0.5506 | 1.465 | 3.88 | 100,242 | 44.9% |
| **3** | Word | 0.1238 | 1.090 | 1.26 | 48,734,695 | 87.6% |
| **3** | Subword | 0.7077 | 1.633 | 4.16 | 388,929 | 29.2% |
| **4** | Word | 0.0464 🏆 | 1.033 | 1.08 | 61,326,075 | 95.4% |
| **4** | Subword | 0.7050 | 1.630 | 3.60 | 1,617,081 | 29.5% |

### Generated Text Samples (Word-based)

Below are text samples generated from each word-based Markov chain model:

**Context Size 1:**

1. `je leta drugi strani gore budizma na sinaju pa mohor bogataj obrobje lokovške planote ima v`
2. `in nkoma kmalu zatem poslala s katero ima toplo hrano kristersson kandiderar till holger in long`
3. `v kombinaciji piracetama ni usmerjen v očetovi smrti ljubljane upravno spada pod konradom auffenstei...`

**Context Size 2:**

1. `je bil dirkalnik konkurenčen in ickx vztrajala da cape financira svoje zadeve anonymous 24 januar go...`
2. `se je letalo prehitelo ekspresni vlak za florido glej tudi seznam francoskih skladateljev g rafael g...`
3. `ki je proučeval razmerja med svojo razgradnjo produktov oksalna kislina je zelo jezen ker je imelo z...`

**Context Size 3:**

1. `glej tudi seznam naselij v črni gori ki upravno spada pod občino čoka slednja pa je del šumadijskega`
2. `pr n št či leta 357 pr n št posvetili cesarju avgustu sklici viri viri ljudstva`
3. `ki ga je po podatkih statističnega urada republike slovenije na pobudo prekmurskega društva general ...`

**Context Size 4:**

1. `glej tudi seznam naselij v srbiji raškega upravnega okraja`
2. `in opombe glej tudi čini slovenske vojske čini ustanovljeni leta čini ukinjeni leta`
3. `opombe glej tudi seznam naselij v srbiji zlatiborskega upravnega okraja`


### Generated Text Samples (Subword-based)

Below are text samples generated from each subword-based Markov chain model:

**Context Size 1:**

1. `_dedna_bhe_ka._(`
2. `ahodnizart;_go_p`
3. `eštske'ijem_z_zo`

**Context Size 2:**

1. `a_gle_nači_zereva`
2. `e_vršče,_cepubrek`
3. `i_pozem_ske_merib`

**Context Size 3:**

1. `je_hočenisoke_dogs`
2. `_je_tren._podatki_`
3. `_potekajo_zgodklon`

**Context Size 4:**

1. `_je_po_znan_pedagog`
2. `_in_renesisteinhart`
3. `_na_polkovniških_od`


### Key Findings

- **Best Predictability:** Context-4 (word) with 95.4% predictability
- **Branching Factor:** Decreases with context size (more deterministic)
- **Memory Trade-off:** Larger contexts require more storage (1,617,081 contexts)
- **Recommendation:** Context-3 or Context-4 for text generation

---
## 4. Vocabulary Analysis

![Zipf's Law](visualizations/zipf_law.png)

![Top Words](visualizations/top20_words.png)

![Coverage Curve](visualizations/vocab_coverage.png)

### Statistics

| Metric | Value |
|--------|-------|
| Vocabulary Size | 897,122 |
| Total Tokens | 72,710,721 |
| Mean Frequency | 81.05 |
| Median Frequency | 4 |
| Frequency Std Dev | 4937.29 |

### Most Common Words

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | je | 2,786,159 |
| 2 | in | 2,040,383 |
| 3 | v | 2,029,005 |
| 4 | na | 1,031,510 |
| 5 | so | 822,699 |
| 6 | se | 719,502 |
| 7 | ki | 679,959 |
| 8 | za | 672,987 |
| 9 | leta | 518,963 |
| 10 | z | 469,417 |

### Least Common Words (from vocabulary)

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | landesgericht | 2 |
| 2 | súkup | 2 |
| 3 | rozec | 2 |
| 4 | malkolma | 2 |
| 5 | sumate | 2 |
| 6 | chamlanga | 2 |
| 7 | presmučala | 2 |
| 8 | luzejem | 2 |
| 9 | edrika | 2 |
| 10 | sigefertovo | 2 |

### Zipf's Law Analysis

| Metric | Value |
|--------|-------|
| Zipf Coefficient | 0.8973 |
| R² (Goodness of Fit) | 0.998517 |
| Adherence Quality | **excellent** |

### Coverage Analysis

| Top N Words | Coverage |
|-------------|----------|
| Top 100 | 30.5% |
| Top 1,000 | 47.9% |
| Top 5,000 | 64.2% |
| Top 10,000 | 71.7% |

### Key Findings

- **Zipf Compliance:** R²=0.9985 indicates excellent adherence to Zipf's law
- **High Frequency Dominance:** Top 100 words cover 30.5% of corpus
- **Long Tail:** 887,122 words needed for remaining 28.3% coverage

---
## 5. Word Embeddings Evaluation

![Embedding Isotropy](visualizations/embedding_isotropy.png)

![Similarity Matrix](visualizations/embedding_similarity.png)

![t-SNE Words](visualizations/tsne_words.png)

![t-SNE Sentences](visualizations/tsne_sentences.png)


### 5.1 Cross-Lingual Alignment

![Alignment Quality](visualizations/embedding_alignment_quality.png)

![Multilingual t-SNE](visualizations/embedding_tsne_multilingual.png)


### 5.2 Model Comparison

| Model | Dimension | Isotropy | Semantic Density | Alignment R@1 | Alignment R@10 |
|-------|-----------|----------|------------------|---------------|----------------|
| **mono_32d** | 32 | 0.7907 | 0.3497 | N/A | N/A |
| **mono_64d** | 64 | 0.7532 | 0.2937 | N/A | N/A |
| **mono_128d** | 128 | 0.6831 | 0.2333 | N/A | N/A |
| **aligned_32d** | 32 | 0.7907 🏆 | 0.3571 | 0.3140 | 0.6660 |
| **aligned_64d** | 64 | 0.7532 | 0.2862 | 0.5440 | 0.8920 |
| **aligned_128d** | 128 | 0.6831 | 0.2351 | 0.6300 | 0.8880 |

### Key Findings

- **Best Isotropy:** aligned_32d with 0.7907 (more uniform distribution)
- **Semantic Density:** Average pairwise similarity of 0.2925. Lower values indicate better semantic separation.
- **Alignment Quality:** Aligned models achieve up to 63.0% R@1 in cross-lingual retrieval.
- **Recommendation:** 128d aligned for best cross-lingual performance

---
## 6.  Morphological Analysis (Experimental)

This section presents an automated morphological analysis derived from the statistical divergence between word-level and subword-level models. By analyzing where subword predictability spikes and where word-level coverage fails, we can infer linguistic structures without supervised data.

### 6.1 Productivity & Complexity

| Metric | Value | Interpretation | Recommendation |
|--------|-------|----------------|----------------|
| Productivity Index | **5.000** | High morphological productivity | Reliable analysis |
| Idiomaticity Gap | **-0.314** | Low formulaic content | - |

### 6.2 Affix Inventory (Productive Units)

These are the most productive prefixes and suffixes identified by sampling the vocabulary for global substitutability patterns. A unit is considered an affix if stripping it leaves a valid stem that appears in other contexts.

#### Productive Prefixes
| Prefix | Examples |
|--------|----------|
| `-s` | samoporjavitev, severvzhodni, svilne |
| `-a` | argonne, aleutskega, aiya |
| `-ma` | mathur, mashel, maerua |
| `-m` | mowbed, migracijah, melodramatične |
| `-k` | kickl, kaškega, koknese |
| `-p` | picramniales, ptujskega, picinus |
| `-b` | brelich, berruguete, bandai |
| `-t` | tragedov, teus, travmatsko |

#### Productive Suffixes
| Suffix | Examples |
|--------|----------|
| `-a` | zimunya, ptujskega, churchosrednja |
| `-i` | zaposliti, gallicani, rurikoviči |
| `-e` | berruguete, melodramatične, svilne |
| `-o` | porfiriato, travmatsko, fajdo |
| `-m` | levičarskem, jožefom, utesnjenem |
| `-ga` | ptujskega, igmanskega, petrarkovega |
| `-s` | picramniales, chlorotis, teus |
| `-ih` | znotrajjetrnih, boucherjevih, pozitivističnih |

### 6.3 Bound Stems (Lexical Roots)

Bound stems are high-frequency subword units that are semantically cohesive but rarely appear as standalone words. These often correspond to the 'core' of a word that requires inflection or derivation to be valid.

| Stem | Cohesion | Substitutability | Examples |
|------|----------|------------------|----------|
| `skeg` | 2.01x | 178 contexts | skega, jskega, muskeg |
| `enje` | 1.57x | 357 contexts | genje, enjeu, ženje |
| `ovez` | 2.40x | 45 contexts | povez, poveza, povezi |
| `orab` | 1.90x | 106 contexts | korab, vorab, porab |
| `ični` | 1.47x | 356 contexts | lični, vični, nični |
| `ijsk` | 1.34x | 565 contexts | bijsk, bijsku, krijska |
| `ranj` | 1.42x | 366 contexts | vranj, ranji, kranj |
| `tičn` | 1.43x | 329 contexts | tičnik, stični, atične |
| `ržav` | 2.01x | 54 contexts | držav, državy, državi |
| `acij` | 1.37x | 302 contexts | lacij, acija, tacij |
| `nske` | 1.34x | 343 contexts | ønske, unske, sanske |
| `avlj` | 1.33x | 320 contexts | javlja, lavlje, kavlja |

### 6.4 Affix Compatibility (Co-occurrence)

This table shows which prefixes and suffixes most frequently co-occur on the same stems, revealing the 'stacking' rules of the language's morphology.

| Prefix | Suffix | Frequency | Examples |
|--------|--------|-----------|----------|
| `-p` | `-a` | 171 words | pumapunkuja, pahmutova |
| `-s` | `-a` | 133 words | soteška, suvalščizna |
| `-p` | `-i` | 120 words | prostorninskimi, prisegami |
| `-p` | `-e` | 100 words | pionirke, priimkie |
| `-k` | `-a` | 97 words | kolegialnega, khandrika |
| `-s` | `-i` | 90 words | sampi, stavkati |
| `-a` | `-a` | 89 words | alkalna, alkibiadesa |
| `-p` | `-o` | 84 words | postkolonialno, plješivico |
| `-b` | `-a` | 83 words | brahmsa, benna |
| `-s` | `-e` | 76 words | stradanje, seksualizirane |

### 6.5 Recursive Morpheme Segmentation

Using **Recursive Hierarchical Substitutability**, we decompose complex words into their constituent morphemes. This approach handles nested affixes (e.g., `prefix-prefix-root-suffix`).

| Word | Suggested Split | Confidence | Stem |
|------|-----------------|------------|------|
| desetstrane | **`desetstr-a-ne`** | 7.5 | `a` |
| globalizirano | **`globalizir-a-no`** | 7.5 | `a` |
| craigallian | **`craigalli-a-n`** | 7.5 | `a` |
| šahrastani | **`šahrast-a-ni`** | 7.5 | `a` |
| sušilnicah | **`sušilnic-a-h`** | 7.5 | `a` |
| sirakuzah | **`sirakuz-a-h`** | 7.5 | `a` |
| grdoselski | **`grdosel-s-ki`** | 7.5 | `s` |
| spremenljivkama | **`spremenljivka-m-a`** | 7.5 | `m` |
| krogličar | **`kroglič-a-r`** | 7.5 | `a` |
| skristaliziral | **`skristalizir-a-l`** | 7.5 | `a` |
| pritajeno | **`pritaj-e-no`** | 7.5 | `e` |
| izhodiščnega | **`izhodišč-ne-ga`** | 7.5 | `ne` |
| lastovsko | **`lastov-s-ko`** | 7.5 | `s` |
| orbeliani | **`orbeli-a-ni`** | 7.5 | `a` |
| igralkodallas | **`igralkodall-a-s`** | 7.5 | `a` |

### 6.6 Linguistic Interpretation

> **Automated Insight:**
The language Slovenian shows high morphological productivity. The subword models are significantly more efficient than word models, suggesting a rich system of affixation or compounding.

---
## 7. Summary & Recommendations

![Performance Dashboard](visualizations/performance_dashboard.png)

### Production Recommendations

| Component | Recommended | Rationale |
|-----------|-------------|-----------|
| Tokenizer | **64k BPE** | Best compression (4.31x) |
| N-gram | **2-gram** | Lowest perplexity (307) |
| Markov | **Context-4** | Highest predictability (95.4%) |
| Embeddings | **100d** | Balanced semantic capture and isotropy |


---
## Appendix: Metrics Glossary & Interpretation Guide

This section provides definitions, intuitions, and guidance for interpreting the metrics used throughout this report.

### Tokenizer Metrics

**Compression Ratio**
> *Definition:* The ratio of characters to tokens (chars/token). Measures how efficiently the tokenizer represents text.
>
> *Intuition:* Higher compression means fewer tokens needed to represent the same text, reducing sequence lengths for downstream models. A 3x compression means ~3 characters per token on average.
>
> *What to seek:* Higher is generally better for efficiency, but extremely high compression may indicate overly aggressive merging that loses morphological information.

**Average Token Length (Fertility)**
> *Definition:* Mean number of characters per token produced by the tokenizer.
>
> *Intuition:* Reflects the granularity of tokenization. Longer tokens capture more context but may struggle with rare words; shorter tokens are more flexible but increase sequence length.
>
> *What to seek:* Balance between 2-5 characters for most languages. Arabic/morphologically-rich languages may benefit from slightly longer tokens.

**Unknown Token Rate (OOV Rate)**
> *Definition:* Percentage of tokens that map to the unknown/UNK token, indicating words the tokenizer cannot represent.
>
> *Intuition:* Lower OOV means better vocabulary coverage. High OOV indicates the tokenizer encounters many unseen character sequences.
>
> *What to seek:* Below 1% is excellent; below 5% is acceptable. BPE tokenizers typically achieve very low OOV due to subword fallback.

### N-gram Model Metrics

**Perplexity**
> *Definition:* Measures how "surprised" the model is by test data. Mathematically: 2^(cross-entropy). Lower values indicate better prediction.
>
> *Intuition:* If perplexity is 100, the model is as uncertain as if choosing uniformly among 100 options at each step. A perplexity of 10 means effectively choosing among 10 equally likely options.
>
> *What to seek:* Lower is better. Perplexity decreases with larger n-grams (more context). Values vary widely by language and corpus size.

**Entropy**
> *Definition:* Average information content (in bits) needed to encode the next token given the context. Related to perplexity: perplexity = 2^entropy.
>
> *Intuition:* High entropy means high uncertainty/randomness; low entropy means predictable patterns. Natural language typically has entropy between 1-4 bits per character.
>
> *What to seek:* Lower entropy indicates more predictable text patterns. Entropy should decrease as n-gram size increases.

**Coverage (Top-K)**
> *Definition:* Percentage of corpus occurrences explained by the top K most frequent n-grams.
>
> *Intuition:* High coverage with few patterns indicates repetitive/formulaic text; low coverage suggests diverse vocabulary usage.
>
> *What to seek:* Depends on use case. For language modeling, moderate coverage (40-60% with top-1000) is typical for natural text.

### Markov Chain Metrics

**Average Entropy**
> *Definition:* Mean entropy across all contexts, measuring average uncertainty in next-word prediction.
>
> *Intuition:* Lower entropy means the model is more confident about what comes next. Context-1 has high entropy (many possible next words); Context-4 has low entropy (few likely continuations).
>
> *What to seek:* Decreasing entropy with larger context sizes. Very low entropy (<0.1) indicates highly deterministic transitions.

**Branching Factor**
> *Definition:* Average number of unique next tokens observed for each context.
>
> *Intuition:* High branching = many possible continuations (flexible but uncertain); low branching = few options (predictable but potentially repetitive).
>
> *What to seek:* Branching factor should decrease with context size. Values near 1.0 indicate nearly deterministic chains.

**Predictability**
> *Definition:* Derived metric: (1 - normalized_entropy) × 100%. Indicates how deterministic the model's predictions are.
>
> *Intuition:* 100% predictability means the next word is always certain; 0% means completely random. Real text falls between these extremes.
>
> *What to seek:* Higher predictability for text generation quality, but too high (>98%) may produce repetitive output.

### Vocabulary & Zipf's Law Metrics

**Zipf's Coefficient**
> *Definition:* The slope of the log-log plot of word frequency vs. rank. Zipf's law predicts this should be approximately -1.
>
> *Intuition:* A coefficient near -1 indicates the corpus follows natural language patterns where a few words are very common and most words are rare.
>
> *What to seek:* Values between -0.8 and -1.2 indicate healthy natural language distribution. Deviations may suggest domain-specific or artificial text.

**R² (Coefficient of Determination)**
> *Definition:* Measures how well the linear fit explains the frequency-rank relationship. Ranges from 0 to 1.
>
> *Intuition:* R² near 1.0 means the data closely follows Zipf's law; lower values indicate deviation from expected word frequency patterns.
>
> *What to seek:* R² > 0.95 is excellent; > 0.99 indicates near-perfect Zipf adherence typical of large natural corpora.

**Vocabulary Coverage**
> *Definition:* Cumulative percentage of corpus tokens accounted for by the top N words.
>
> *Intuition:* Shows how concentrated word usage is. If top-100 words cover 50% of text, the corpus relies heavily on common words.
>
> *What to seek:* Top-100 covering 30-50% is typical. Higher coverage indicates more repetitive text; lower suggests richer vocabulary.

### Word Embedding Metrics

**Isotropy**
> *Definition:* Measures how uniformly distributed vectors are in the embedding space. Computed as the ratio of minimum to maximum singular values.
>
> *Intuition:* High isotropy (near 1.0) means vectors spread evenly in all directions; low isotropy means vectors cluster in certain directions, reducing expressiveness.
>
> *What to seek:* Higher isotropy generally indicates better-quality embeddings. Values > 0.1 are reasonable; > 0.3 is good. Lower-dimensional embeddings tend to have higher isotropy.

**Average Norm**
> *Definition:* Mean magnitude (L2 norm) of word vectors in the embedding space.
>
> *Intuition:* Indicates the typical "length" of vectors. Consistent norms suggest stable training; high variance may indicate some words are undertrained.
>
> *What to seek:* Relatively consistent norms across models. The absolute value matters less than consistency (low std deviation).

**Cosine Similarity**
> *Definition:* Measures angular similarity between vectors, ranging from -1 (opposite) to 1 (identical direction).
>
> *Intuition:* Words with similar meanings should have high cosine similarity. This is the standard metric for semantic relatedness in embeddings.
>
> *What to seek:* Semantically related words should score > 0.5; unrelated words should be near 0. Synonyms often score > 0.7.

**t-SNE Visualization**
> *Definition:* t-Distributed Stochastic Neighbor Embedding - a dimensionality reduction technique that preserves local structure for visualization.
>
> *Intuition:* Clusters in t-SNE plots indicate groups of semantically related words. Spread indicates vocabulary diversity; tight clusters suggest semantic coherence.
>
> *What to seek:* Meaningful clusters (e.g., numbers together, verbs together). Avoid over-interpreting distances - t-SNE preserves local, not global, structure.

### General Interpretation Guidelines

1. **Compare within model families:** Metrics are most meaningful when comparing models of the same type (e.g., 8k vs 64k tokenizer).
2. **Consider trade-offs:** Better performance on one metric often comes at the cost of another (e.g., compression vs. OOV rate).
3. **Context matters:** Optimal values depend on downstream tasks. Text generation may prioritize different metrics than classification.
4. **Corpus influence:** All metrics are influenced by corpus characteristics. Wikipedia text differs from social media or literature.
5. **Language-specific patterns:** Morphologically rich languages (like Arabic) may show different optimal ranges than analytic languages.


### Visualizations Index

| Visualization | Description |
|---------------|-------------|
| Tokenizer Compression | Compression ratios by vocabulary size |
| Tokenizer Fertility | Average token length by vocabulary |
| Tokenizer OOV | Unknown token rates |
| Tokenizer Total Tokens | Total tokens by vocabulary |
| N-gram Perplexity | Perplexity by n-gram size |
| N-gram Entropy | Entropy by n-gram size |
| N-gram Coverage | Top pattern coverage |
| N-gram Unique | Unique n-gram counts |
| Markov Entropy | Entropy by context size |
| Markov Branching | Branching factor by context |
| Markov Contexts | Unique context counts |
| Zipf's Law | Frequency-rank distribution with fit |
| Vocab Frequency | Word frequency distribution |
| Top 20 Words | Most frequent words |
| Vocab Coverage | Cumulative coverage curve |
| Embedding Isotropy | Vector space uniformity |
| Embedding Norms | Vector magnitude distribution |
| Embedding Similarity | Word similarity heatmap |
| Nearest Neighbors | Similar words for key terms |
| t-SNE Words | 2D word embedding visualization |
| t-SNE Sentences | 2D sentence embedding visualization |
| Position Encoding | Encoding method comparison |
| Model Sizes | Storage requirements |
| Performance Dashboard | Comprehensive performance overview |

---
## About This Project

### Data Source

Models trained on [wikipedia-monthly](https://huggingface.co/datasets/omarkamali/wikipedia-monthly) - a monthly snapshot of Wikipedia articles across 300+ languages.

### Project

A project by **[Wikilangs](https://wikilangs.org)** - Open-source NLP models for every Wikipedia language.

### Maintainer

[Omar Kamali](https://omarkamali.com) - [Omneity Labs](https://omneitylabs.com)

### Citation

If you use these models in your research, please cite:

```bibtex
@misc{wikilangs2025,
  author = {Kamali, Omar},
  title = {Wikilangs: Open NLP Models for Wikipedia Languages},
  year = {2025},
  doi = {10.5281/zenodo.18073153},
  publisher = {Zenodo},
  url = {https://huggingface.co/wikilangs}
  institution = {Omneity Labs}
}
```

### License

MIT License - Free for academic and commercial use.

### Links

- 🌐 Website: [wikilangs.org](https://wikilangs.org)
- 🤗 Models: [huggingface.co/wikilangs](https://huggingface.co/wikilangs)
- 📊 Data: [wikipedia-monthly](https://huggingface.co/datasets/omarkamali/wikipedia-monthly)
- 👤 Author: [Omar Kamali](https://huggingface.co/omarkamali)
- 🤝 Sponsor: [Featherless AI](https://featherless.ai)
---
*Generated by Wikilangs Models Pipeline*

*Report Date: 2026-01-17 08:23:12*